IoTA: IoT Automated SIP-based Emergency Call Triggering System for generaleHealth purposes.

Foteini Andriopoulou,Theofanis OrphanoudakisSchool of Science and Technology

Hellenic Open UniversityPatras, 26335, Greece

Email: fandriopoulou@eap.gr,fanis@eap.gr

Tasos DagiuklasSchool of Engineering, Computer Science and Informatics

London South Bank University103 Borough Road,London SE1 0AA

Email: tdagiuklas@lsbu.ac.uk

Abstract—The expansion of Internet of Things (IoT) and theevolution in communication technologies have enabled homes,cars even whole cities to be network connected. However,during an emergency incident, IoT devices have not been usedto trigger emergency calls directly to healthcare providersmainly due to their constrained capabilities and lack of sup-port session-oriented communications. Moreover, emergencyservices are currently offered by public safety stakeholdersthat do not support call triggering by IoT devices. This paperproposes IoTA framework which enables IoT devices to gen-erate automatically emergency calls and support bi-directionalcommunication sessions between healthcare providers and endusers. The IoTA framework incorporates intelligent algorithmsfor processing and evaluating emergency events from variousdevices and performs emergency calls immediately after theoccurrence of an event. The healthcare providers can interactwith the IoTA framework requesting continuous real-timesensor data. A prototype implementation and initial evaluationresults are presented as a proof of concept for people sufferingfrom diverse chronic diseases. Experimental results have shownthat the proposed framework can be considered as a promisingsolution for detecting, reporting emergency events, eliminatingthe hoax calls and responding swiftly saving lives.

Keywords-emergency calls; Internet of Things; IoT; sensordevices; IP communication session; SIP based calls; eHealth;seizure

I. INTRODUCTION

As every device and appliance can potentially be con-nected to the Internet, Internet of Things (IoT) has gainedthe attention of the industry and research domain [1]. Anumber of research projects headed by different institutesand industry have been focusing on IoT application andservice provision for health monitoring [2].

In the domain of smart homes, numerous sensor devicesare used to support Ambient Assisted Living (AAL) suchas motion and fall detectors, smoke alarms, bed detectorsand others [3]. These devices are usually used to moni-toring events related to the user’s healthcare condition orfalls and notifying a caregiver or a member of the familyabout an emergency incident. Thus, the receiver of the alertservice is responsible to trigger the emergency call andprovide information regarding the emergency event such as

the symptoms, the cause, location, and the severity of theincident, information that most of the times does not evenknow. Moreover, the intervention of the intermediate receiverincreases the response time that may lead to increased costsdue to hospitalization or unpredictable loses of human life.In emergency cases, the time between the occurrence ofthe event and the service provision is crucial and shouldbe eliminated. Enabling IoT and sensor devices to generateautomatically emergency calls is a challenging issue forresponding in an efficient thus accurate way to emergencyincidents and avoiding unreasonable waste of time.

In this paper, an emergency framework (IoTA framework)is proposed that meets all the aforementioned challenges bygenerating automatically emergency calls to the appropriatehealthcare provider and/or public safety answering point(PSAP). IoTA framework incorporates sensors that monitorthe vital-signs and the body position of people suffering fromdiverse chronic diseases such as seizures in order to detectabnormal events. It incorporates outlier detection algorithmsfor filtering the aggregated data and identifying if the ab-normal event occurs due to an emergency event. Wheneveran alarm is triggered due to an abnormal event, it usesintelligent decision algorithms in order to distinguish its typeand severity. Then, it forwards the sensor data to a SIP client(i.e. SIP phone) that establishes an emergency call with theappropriate healthcare provider supplying information aboutthe location of the event, sensor data, video and voice. Usingdecision algorithms based on the set of the available sensordevices eliminates the hoax calls triggered by a single sensor.Moreover, it supports mechanisms that enable the call takerof the emergency call to request continuous sensor data usinga bi-directional communication with the IoT devices.

A prototype application has been implemented as a proofof concept in order to demonstrate the basic features andthe effectiveness of the proposed system. The case studyused for evaluation purposes is based on a real-time scenariofor monitoring patients suffering from seizures and needhelp immediately in order to avoid injuries. However, theproposed solution can be easily applicable to many IoT-based solutions such as ecall systems or smart cities in order

to perform emergency calls.The rest of the paper is organized as follows: in section II

is presented the related work. The proposed IoTA frameworkarchitecture is presented in section III. In Section IV and V,the implementation details and a qualitative evaluation ofthe proposed IoTA framework are presented and discussedrespectively. Finally, Section VI concludes the whole paper.

II. RELATED WORK

The World Health Organisation (WHO) [4] refers thatseizure episodes are a result of excessive electrical dis-charges in a group of brain cells. Seizures can vary infrequency and in the part of the brain the disturbance firststarts. Temporary symptoms occur, such as loss of awarenessor consciousness, and disturbances of movement, sensation,mood, or other cognitive functions. One seizure does notsignify epilepsy, a number of conditions can imitate seizuressuch as mini-strokes, hypoglycemia, panic attacks, etc. Peo-ple with seizures tend to have more physical problems (suchas fractures and bruising from injuries related to seizures)[4].Studies report that providing emergency services in an accu-rate way is crucial and can reduce the impact of accidents.For a noticeable reduction of the response time, is mandatorythe fast and accurate detection of the emergency event andthe reporting to the appropriate healthcare provider in orderto receive immediately emergency services.

In the healthcare domain, IoT devices have been used inorder to detect or prevent seizures. Many algorithms andintelligent algorithms have been proposed that evaluate theelectroencephalography (EEC) data received from the user’sscalp, electromyography (EMG) or electrocardiogram (ECG)data. Cook et al. [5] proposed an intracranial implanteddevice used to collect EEG data on the cortical surface. Theelectrode leads were connected to an implanted telemetryunit, which wirelessly transmitted data to an external, hand-held personal advisory device. The external device receivedthe EEG data, applied an algorithm to them, and displayedthe resultant information to the user’s device. Authors in [6],[7] and [8] have proposed the use of accelerometers eitherwrist-worn or limb and chest worn for detecting general-ized tonicclonic seizures. Different mechanisms have beenimplemented using linear thresholds, standard deviations,short-time Fourier transform, pattern recognition algorithmsor time-domain and frequency-domain based algorithms inorder to identify the existence of a seizure event [9]. In allthese works, once a seizure was detected by the algorithm,the control unit triggered an alarm calling for help. Thealarm was sent to the caretakers.

Seizures can be unpredictable. Some may start with minorsymptoms, and then lead to a fall or a loss of consciousness.In this context, fall detection algorithms will be useful inorder to detect the health status of the person experience aseizure. In the bibliography, there are various fall detectionsystems and classification algorithms have been proposed

and evaluated in order to increase the accuracy of the detec-tion or prevent a fall. Yazar, Erden and Cetin [10] proposeda low-cost and privacy-friendly fall detection system forAAL using vibration and passive infrared (PIR) sensors.Yanuz et al [11] proposed an enhanced smartphone-basedfall detection system with location information using GoogleMaps about the person experienced the fall. When a fallis detected, the available GPS interface on the smartphonetransmits the notification of the fall incident to the caregivervia SMS, email and Twitter messages. Habib et al [12]proposed a smartphone-based fall detection and preventionsystem for fall management. Combinations of alarm systemsand graphical user interface are used for collecting thefeedback of the user. If the user does not respond withinthat time, the system will consider the event as a fall andgenerate several types of notifications to seek help fromcaregivers (e.g. SMS, MMS, email,twitter messaging, etc).Sposaro and Tyson [13] proposed a similar system for falldetection using common commercially available electronicdevices to both detect the fall and alert authorities.If a fall issuspected a notification is raised requiring the users responseor a confirmation of user’s social contact about the event.

The aforementioned works have been focused on detect-ing either seizures or falls using a series of sensor deviceswearable or incorporated into smartphones and notifyinguser’s caregivers and relatives. The caregivers are responsi-ble for triggering the emergency call systems and reportinginformation regarding the emergency event. The existenceof reporting to the caregivers an emergency call increasesthe response time that may lead to unpredictable loses ofhuman life. To the best of our knowledge, the work ofCirani, Picone and Veltri [14], was the first that enabledIoT devices to perform SIP-based communication sessionsin a lightweight and standard fashion. Then, NEXES [15]project proposed the integration of the sensor devices usingthe MSD format in order to establish automatically SIP-based emergency calls.

In this paper, we propose IoTA framework that meetsall the aforementioned challenges by generating automati-cally emergency calls to the appropriate healthcare provider.IoTA framework is not another framework that promisesto prevent seizures but it is totally differentiated fromthe above mentioned works in three basic ways. Firstly,it correlates data from multiple sensors that identify theexistence of an emergency event due to seizure or fall andit can support the user during and after the emergencyevent triggering automatically healthcare providers. More-over, it performs communication sessions directly with thehealthcare providers forwarding the sensor data to a SIP-based client that establishes an emergency call with them.Healthcare providers can have access and request continuousdata transmission in order to assist the patient in an accurateway. Finally, contrary to the CoSIP work [14], IoTA isindependent of the application protocol that sensors use to

transmit data.

III. IOTA FRAMEWORK ARCHITECTURE

The IoTA Framework is located at the caller’s side. Itcomprises the sensor platform, the sensor data interface andthe SIP client as presented in Fig.1. The caller’s side can bealso equipped with various appliances (i.e. laptops, tablets,TV sets, microphones, speakers), environmental sensors (i.e.humidity, smoke detector, motion) and healthcare monitoringsensors (i.e. pulsioximeter, thermometer, ECG). These sen-sors monitor the environmental conditions and bio-signalsthat indicate the health status of the caller.

Figure 1. The IoTA Framework architecture overview.

The sensor platform is responsible for aggregating en-vironmental, medical data and the user’s position frommultiple sensors. These data are stored either to an internalSensorDatabase for processing or to an external cloud baseon the healthcare provider’s side. Sensor platform incorpo-rates multiple algorithms for processing and evaluating thesensor data so as to detect abnormal events and identify theirtype, if it is an outlier value or an emergency event. Asan emergency event is detected, the sensor platform createsan alert and sends the sensor data in an XML-format tothe SIP client i.e. SIP-based VoIP application - SIP phone)establishing a communication session with the appropriatehealthcare provider.

The SIP Client is a key component for emergency servicesas it supports location information and routing to the appro-priate healthcare provider. It is enhanced in order to useUniform Resource Names (URNs) identifiers for IP-basedemergency calls and support SIP SUBSCRIBE/NOTIFYmessages. The URN service allows the identification ofan emergency service to be triggered for each type ofemergency event (i.e. ambulance, emergency room, etc)overlooking the need to manually search for the providerof that service. The implementation of the URN in the SIPclient enables accurate report to the appropriate providereliminating time delays due to manual search for theprovider of that service. Moreover, in the IoTA Frameworkthe SIP client is envisioned to have an additional interface tosensor platforms so as to provide enhanced functionalities byreporting the status of the emergency caller. In this context,

the SIP client supports the SIP SUBSCRIBE/NOTIFY mes-sage [16] providing a bi-directional communication betweenthe citizens, their IoT devices and the healthcare providers.This di-directional communication enables the call taker tobe informed in an accurate way and request continuous real-time sensor data streaming in order to detect the emergencyevent, estimate its severity and respond swiftly and effec-tively, saving lives.

The sensor data interface has been developed as a clientserver model in order to enable the communication be-tween the sensor platform and the SIP client. The interfaceaccepts the critical data in XML format. Sensor ModelLanguage (SensorML) [17] has been selected for standardbased, unified sensor data representation model. Then, itencapsulates them in the body of SIP messages and forwardsthem to the healthcare provider through normal SIP routing.This approach is in conformity with already standardizedsolutions, like PIDF-LO (RFC 513943), and thus can beefficiently adapted to existing solutions.

A. Sensor Platform

The sensor platform is composed of the following com-ponents (Fig. 2):

Figure 2. The components of the sensor platform architecture.

• Data Acquisition Module: it provides two interfaces forreceiving data. The former enables real time sensor dataaggregation and storing and the latter one allows dataretrieval for updating purposes.

• Decision Making Module: it processes and evaluatesreal time data received from medical sensors in order todetect emergency events. As soon as an abnormal eventis detected, it implements algorithms to detect eitherpossible sensor failures or actual emergency incidents,creates an alarm and forwards the sensor data regardingthe emergency event to the UDP Client component.

• UDP Client: This component creates an XML file thatcontains the sensor data received from the DecisionMaking Module or Data Acquisition Module compo-nent and sends it over UDP/IP protocol to a SIP client.

• UDP Server: This component provides an interfaceto the SIP client that allows the PSAP respondersto request updated real time data of the caller thatinitiated the emergency call. The UDP Server receivesSIP messages.

B. SIP Client

The SIP client consists of the following components(Fig.3):

• UrnMapping: This sub-component performs the trans-lation of emergency number, such as 112, to the corre-sponding URN service identifier (e.g., urn:service:sos).

• SensorCommunication: This component implementsthe communication interface to the sensor platform. It isbased on the MESSAGE SIP method [18] and exploitsUDP as transport. It receives an alert in the case of anabnormal medical event and triggers the InitiateEmer-gencyCall. Whenever an update of the sensor values isrequired, the component is also able to make a requestto the sensor platform and forward the collected datato the SensorUpdateEvent component.

• InitiateEmergencyCall: It initiates an enhanced emer-gency call. In this case, the medical data are included inthe body of the INVITE message in SensorML format[17]. It also includes the provided URN identifier in theTo: field of the INVITE message.

• SensorUpdateEvent: This component supports the up-date of the sensor values during an emergency event atthe client-side. For that purpose the SIP SUBSCRIBE-NOTIFY framework [16] is utilized, in order to al-low the involved parts to get near real-time updatedvalues of the callers sensor data (i.e. vital signs andenvironmental data). The SIP specific event notificationframework allows the entities of the network (e.g.,PSAP, first responders) to subscribe to the state of otherentities, which in return can send notifications whentheir state changes.

Figure 3. The components of the SIP Client.

IV. IMPLEMENTATION

A. System SetUp

A prototype application was developed as a proof ofconcept for IoTA Framework. The application includes twomain components: the sensor platform and the SIP Client.

The sensor platform is developed upon a low powersingle-board computing device (Raspberry Pi2) running De-bian operating system and using Ethernet or IEEE 802.11wireless connection to the Internet. The software of thesensor platform is developed in C++ multithreading pro-gramming model. The software that collects data from eachsensor is implemented and runs as an independent thread

enabling fast execution of the outlier detection and decisionprocesses and optimizing resource consumption, latency andresponsiveness. The sensor data are stored locally to adatabase located at Pi for temporary usage and to a clouddatabase located to the healthcare provider for permanentstorage. The InfluxDB database [19] was used in both cases.An open source metric analytics and visualization tool,Grafana [20] was used in both user and healthcare provider’send in order to visualize time series data from sensors.

The SIP Client is implemented based on the open-source Voice Over IP phone, Linphone [21]. SIP Clientwas extended in order to encapsulate information regardingthe location and the sensor data. SIP Client retrieves theuser’s location from location servers. The LoST (Locationto Service Translation) protocol [22] which maps URNsand location information to one or more service URLs isused to retrieve the location of the appropriate healthcareprovider. SIP Client retrieves the location by the providersusing a PIDF-LO XML payload and encapsulates it intothe SIP INVITE message for initiating an emergency call.Moreover, similar to the location data, SIP Client receivesthe aggregated bio-signals from sensors into an XML-formatwhich is encapsulated to the SIP INVITE message body.SIP Client developed in C, runs on Ubuntu 14.04 operatingsystem.

B. Decision Process

The decision process is divided into two discrete phases(Fig. 4). In the first phase, the readings of the sensors areevaluated in order to prune the outlier values. In the secondphase of the decision process, a hybrid emergency detectionalgorithm is proposed that detects falls and correlates bio-signs data in order to detect emergency events. Wheneveran emergency event is identified, sensor data are sent to theSIP Client triggering an emergency call.

1) Outlier Detection: IoT and sensory devices send dataand vital-signs to the sensor platform. The software ofeach sensor acquires and processes the incoming data ofthe sensor at a rate of 1 measurement per second. Foreach sensor, predefined threshold values are used in orderto detect abnormal events. As normal values we definevalues that their distance from the other time-series datais greater than the threshold margin. When an abnormalevent is detected, a time-series outlier detection algorithmis performed in order to discern conditions between a faultysensor reading and a patient entering into an emergencystate. The outlier detection algorithm uses the median valueof an x-time-series data to determine whether the currentsensor reading is an outlier (Algorithm 1).

2) Hybrid Emergency Detection Algorithm: If the abnor-mal value is not labeled as outlier, a three-phase hybridemergency detection algorithm is performed distinguishingthe type (fall, seizure or both) and the severity of theemergency. The one-phase algorithm evaluates data from

Figure 4. The flow diagram of the decision process.

Algorithm 1 Outlier Detection Algorithm for sensor dataif the y-value of sensor is out of the predefined maximum-minimum threshold range then

calculate the difference of the values of the x-seriescalculate median value of the x-series datacalculate median difference of the x-series valuescalculate threshold difference ← maxThreshold −minThresholdif |currentV alue−median| >|threshold difference±median difference|then

label as outlier & increase NumOutliersend if

end if

accelerometer and gyroscope for detecting falls. We usedthe algorithm proposed by [23] due to the high accuracy(99.382%) in a timely manner that performs in contraryto other more complex algorithms. In the second-phase,motivated by the results presented in [24], we correlatedbio-signals from the medical sensors (ECG, blood pressure,pulsioximeter, body thermometer, airflow) using J48 classifi-cation algorithm in order to detect seizure and/or stroke [25].The J48 classification algorithm generates decision trees, thenodes of which evaluate the existence or significance of aseizure or stroke event. A decision-tree model is built byanalyzing training data and the model is used to classifyunseen data. In our model, the tree nodes are the monitoredsensor data and the leaf nodes are the class (seizure, stroke,normal). Then, using a rule-based algorithm we decide for

the generation of the emergency alarm (Algorithm 2) .

Algorithm 2 Decision making algorithm identifying emer-gency events

if SensorValue == ABNORMAL ∥ NumOutliers > 10then

check if a fall occursif FALL == TRUE then{check bio signals}if (SEIZURE ∥ STROKE) == TRUE then

initiate an alarm sos.ambulanceelse

if FALLEN > 2Minutes theninitiate an alarm sos.ambulance

end ifend if

elsecheck if a seizure or stroke occursif (SEIZURE ∥ STROKE) == TRUE then

initiate an alarm sos.ambulanceend if

end ifend if

C. Communication EstablishmentThe sensor platform aggregates, processes and evaluates

the sensor data. Whenever an emergency incident is detected,the sensor platform device sends the sensor data related tothe emergency event using the SensorML [17] format to theSIP client via the UDP Client. The combined sensor data aresent as part of the SIP INVITE message via the Linphoneapplication running on user’s laptop or smartphone. Therecipient of the emergency call is able to request continuousupdating of the sensor data. To support this, the sensor plat-form incorporates a UDP Server. The UDP Server runningon the sensor platform receives messages from the SIP-based application, parses the header, stores the destinationand source IP addresses as well as the CallID and thetime interval dictated by the healthcare provider to receiveupdated sensor data. Then, it forwards to the SIP client theupdated sensors data until it receives the message to stopforwarding. Fig. 5 presents the communication establishmentbetween the sensor platform and the SIP Client.

Figure 5. The sequence diagram of the internal communication establish-ment of IoTA Framework.

Through the SIP client, each emergency event enters thesystem and joins the appropriate healthcare according to itstype of alarm and location. The SIP client is responsible forestablishing and maintaining the SIP communication sessionand updating the sensor data to the healthcare side throughSIP SUBSCRIBER/NOTIFY method. Whenever SIP Clientis triggered by the sensor platform, SIP CLient retrievesthe user’s location from location service providers (DHCP,HELD, LLDP-MED, Android Location Services). SIP Clientreceives the location information in a PIDF-LO XML formatand encapsulates it in a SIP INVITE. SIP INVITE messageencapsulates to its payload also data from the sensors inan XML format. SIP INVITE message and is sent to theSIP Proxy. SIP proxy is extended to support emergencyservices recognition and routing. It receives a SIP call witha URN indicating the emergency service requested (i.e.ambulance) requests from a location service to determinethe appropriate healthcare provider to route the call. Thenthe call is forwarded to the SIP Client of the healthcareprovider (the recipient of the call). For the event notification,the SIP specific event notification framework is used so as toallow emergency stakeholders (e.g. healthcare providers, firstresponders) to subscribe to the state of other entities, whichin return can send notifications when their state changes.Fig. 6 presents the communication establishment betweenthe SIP Client and the healthcare provider.

Figure 6. The sequence diagram of the inter communication establishmentof IoTA Framework.

D. System Demonstration

In Fig. 7 is presented the screenshot during the automat-ically emergency triggering process. The IoTA frameworkinitiated an automated triggering call due to a fall that lastedmore than 2 minutes. A multimedia communication sessionis established with the healthcare and the sensor data as wellas the location information are presented in a user-friendlyGraphical User Interface (GUI).

Figure 7. Screenshot of IoTA framework triggering an emergency call tothe healthcare Provider.

V. EVALUATION

In order to support patients suffering from diverse chronicdiseases that may experience a seizure (i.e. epilepsy, is-chemic stroke) a set of lightweight sensors are providedto the user. These sensors include body thermometer, pul-sioxemeter, ECG, blood pressure, airflow and galvanic Skinresponse sensors. A tri-axial accelerometer ADXL345 anda gyroscope ITG3200 were used in order to measure ac-celeration and angular velocity for detecting falls. Theaccelerometer/gyroscope is packed into a chest belt foroptimum location retrieval. All the sensors are connecteddirectly to the sensor platform. The sensor platform is awireless portable device carried into the user’s pocket.

The experiment is performed on data from 15 elderlyhealthy and non-healthy subjects (6 men and 9 women,aged from 60 to 75 years). The data emulate subjects thatmonitor their healthcare status using the medical sensorsand simultaneously perform their daily activities includingwalking, ascending/descending stairs, sitting, standing orsleeping.

In order to evaluate the effectiveness of the proposedsystem, a logging/auditing service was implemented on thesensor platform. It is activated when a sensor value isreceived and creates log files reporting the number of theabnormal values before and after the performance of theoutlier detection and the decision algorithms are countedso as to identify the accuracy of of IoTA framework. Theaccuracy and the error hereafter hoax calls are defined as:

Accuracy = TPAlarmTPAlarm+FNAlarm

HoaxCalls = TNAlarmTNAlarm+FPAlarm

where:• TPAlarm: it is the number of true positive alarms by

means of emergency event occurred and detected byIoTA.

• FPAlarm: it is the number of false positive alarmsmeaning that IoTA detects an emergency event whichdoes not occur.

• TPAlarm: it is the number of true negative alarmsmeaning that there is not any emergency event andIoTA detects it as normal (non-emergency)event.

• FNAlarm: it is the number of false negative alarmsmeaning that an emergency event occurs and IoTA doesnot detect it.

By observing the results shown in Table I, it comes outthat incorporating multiple sensors and decision schemeshas reduced the number of hoax calls. This is due to thefact that inherently, the data retrieved by a single sensorare not reliable to identify emergency events. It should bementioned that the accuracy of the fall, seizure or strokedetection algorithm (presented in Table I) is related to theireffectiveness to identify an emergency event. This is totallydifferent from their effectiveness to identify the event forwhich they are implemented. For instance, during a seizurethe person loses his consciousness and falls. In this case,the fall and seizure detection algorithm would trigger analarm. However, if the seizure lasts less than 5 minutes andthe fallen person is not injured, there is no need to triggeran emergency call. The proposed decision algorithm takesinto consideration these parameters leading to a more robustand effective emergency detection system (accuracy 82.86%)eliminating the hoax calls which are estimated to 17.14%.

Table ITHE ACCURACY OF IOTA FRAMEWORK TO DETECT ABNORMAL EVENTS

AND TRIGGER A CALL.

Abnormal events identified Accuracy % HoaxCalls(%)

Using the outlier algorithm (Algorithm 1) 20.68 79.32Using fall detection algorithm 45.21 64.79Using seizure and stroke detection algo-rithms

63.83 36.17

Using the proposed decision algorithm (Al-gorithms 1 & 2)

82.86 17.14

Moreover, the auditing service creates log files report-ing the processing time the oultier and decision algorithmrequire in order to identify an emergency event. Table IIpresents the average times in order to evaluate the receivedsensor data, perform decision algorithms and send the datato a pre-defined format ([17]) to the SIP client. The averagetime to identify if an abnormal value is an outlier, isestimated between 1 and 2 sec. This is due to the fact thatthe outlier detection algorithm verifies that an outlier existsafter performing the same process for the next incomingvalue. Each sensor is implemented to send data to the sensorplatform under a predefined time that varies from 1 to 2seconds. As a consequence, when the second value arrivesat the sensor platform, it is processed. The hybrid emergencydetection algorithm performs an internal evaluation andverification process estimating the existence of an emergencycase. The average total time is estimated at about 11.5 secs.

Table IIEVALUATION RESULTS OF SENSOR PLATFORM

Process Evaluation AverageTime (s)

Outlier Algorithm processing 1.5Hybrid emergency detection Algorithm pro-cessing

10

Total Response 11.5

For the initiation of a SIP session, the signaling cost forthe extended SIP INVITE message encapsulated with datafrom six different sensors is equal to 1128 bytes while theSIP NOTIFY is equal to 582 bytes. As the number of sensorsis increased, the length of the SIP INVITE and NOTIFYmessage is increased. Considering the CoSIP approach [14]that enables each sensor to generate an alarm, for a smallnumber of sensors, only two, the extended SIP INVITEmessage overhead (670 bytes) is similar to this provided byCoSIP (2sensors ∗ 311CoSIPbytes/sensor = 622bytes)[14]. However, for multiple sensors the extended SIP IN-VITE yields the CoSIP with a compression ratio above 0.5(Table III). The signaling cost of the SIP communication andthe average time required to establish a communication ses-sion with the healthcare provider is planned to be evaluatedin future extensions of this work.

Table IIICOMPARISON BETWEEN EXTENDED SIP AND COSIP MESSAGE

OVERHEAD FOR SIP INVITE MESSAGE.

Number ofSensors

extended SIP(bytes)

CoSIP (bytes) compressionratio

2 670 622 1.0774 937 1244 0.7536 1128 1866 0.605

VI. CONCLUSION

In this paper, IoTA framework is proposed that en-ables IoT devices to generate automatically emergency callsto healthcare providers supporting people suffering fromchronic diseases that live independently. IoTA frameworkaggregates and evaluates sensor data in order to detectabnormal events. It incorporates outlier detection algorithmsand decision making algorithms that identify if the ab-normal event is an emergency event or not. Whenever analarm is triggered due to an emergency event, the IoTAframework forwards the sensor data to a SIP client thatis responsible for establishing emergency calls with theappropriate healthcare provider providing a set of locationand sensor data information related to the incident. TheSIP client is enhanced in order to use the URN serviceidentifier for IP-based emergency calls and support SIPSUBSCRIBE/NOTIFY messages. In this context, the calltaker can request continuous real-time sensor data using abi-directional communication with the IoT devices.

A prototype application was implemented as a proof ofconcept in order to demonstrate the basic features and the

effectiveness of the proposed system supporting patientsduring and after a seizure. The evaluation results illus-trate that the IoTA Framework can be considered as apromising solution for detecting, reporting emergency eventsand responding swiftly saving lives. The proposed solutioncan be easily incorporated to many IoT-based solutions soas to perform emergency calls and respond effectively inemergency incidents such as ecall systems or smart cities.

ACKNOWLEDGMENT

This work is supported by H2020 European FrameworkProgramme under the EMYNOS - nExt generation eMer-gencY commuNicatiOnS project.

REFERENCES

[1] L. Atzori, A. Iera, and G. Morabito, “The internet of things: Asurvey,” Computer networks, vol. 54, no. 15, pp. 2787–2805,2010.

[2] N. Bui and M. Zorzi, “Health care applications: a solutionbased on the internet of things,” in Proceedings of the 4thInternational Symposium on Applied Sciences in Biomedicaland Communication Technologies. ACM, 2011, p. 131.

[3] A. Pantelopoulos and N. G. Bourbakis, “A survey on wearablesensor-based systems for health monitoring and prognosis,”IEEE Transactions on Systems, Man, and Cybernetics, PartC (Applications and Reviews), vol. 40, no. 1, pp. 1–12, 2010.

[4] (02/2017) Epilepsy. [Online]. Available: http://www.who.int/mediacentre/factsheets/fs999/en/

[5] M. J. Cook, T. J. O’Brien, S. F. Berkovic, M. Murphy,A. Morokoff, G. Fabinyi, W. D’Souza, R. Yerra, J. Archer,L. Litewka et al., “Prediction of seizure likelihood witha long-term, implanted seizure advisory system in patientswith drug-resistant epilepsy: a first-in-man study,” The LancetNeurology, vol. 12, no. 6, pp. 563–571, 2013.

[6] J. Lockman, R. S. Fisher, and D. M. Olson, “Detection ofseizure-like movements using a wrist accelerometer,” Epilepsy& Behavior, vol. 20, no. 4, pp. 638–641, 2011.

[7] U. Kramer, S. Kipervasser, A. Shlitner, and R. Kuzniecky,“A novel portable seizure detection alarm system: preliminaryresults,” Journal of Clinical Neurophysiology, vol. 28, no. 1,pp. 36–38, 2011.

[8] S. Beniczky, T. Polster, T. W. Kjaer, and H. Hjalgrim, “Detec-tion of generalized tonic–clonic seizures by a wireless wristaccelerometer: a prospective, multicenter study,” Epilepsia,vol. 54, no. 4, 2013.

[9] S. Ramgopal, S. Thome-Souza, M. Jackson, N. E. Kadish,I. S. Fernandez, J. Klehm, W. Bosl, C. Reinsberger,S. Schachter, and T. Loddenkemper, “Seizure detection,seizure prediction, and closed-loop warning systems inepilepsy,” Epilepsy & behavior, vol. 37, pp. 291–307, 2014.

[10] A. Yazar, F. Erden, and A. E. Cetin, “Multi-sensor ambientassisted living system for fall detection,” in Proceedings ofthe IEEE International Conference on Acoustics, Speech, andSignal Processing (ICASSP14). Citeseer, 2014, pp. 1–3.

[11] G. Yavuz, M. Kocak, G. Ergun, H. O. Alemdar, H. Yalcin,O. D. Incel, and C. Ersoy, “A smartphone based fall detectorwith online location support,” in International Workshop onSensing for App Phones; Zurich, Switzerland, 2010, pp. 31–35.

[12] M. A. Habib, M. S. Mohktar, S. B. Kamaruzzaman, K. S.Lim, T. M. Pin, and F. Ibrahim, “Smartphone-based solutionsfor fall detection and prevention: challenges and open issues,”Sensors, vol. 14, no. 4, pp. 7181–7208, 2014.

[13] F. Sposaro and G. Tyson, “ifall: an android application forfall monitoring and response,” in Engineering in Medicineand Biology Society, 2009. EMBC 2009. Annual InternationalConference of the IEEE. IEEE, 2009, pp. 6119–6122.

[14] S. Cirani, M. Picone, and L. Veltri, “Cosip: a constrainedsession initiation protocol for the internet of things,” in Euro-pean Conference on Service-Oriented and Cloud Computing.Springer, 2013, pp. 13–24.

[15] Nexes: Next generation emergency services. [Online].Available: http://nexes.eu.

[16] A. Roach, “Sip-specific event notification,” 2012.

[17] M. Botts, A. Robin, J. Greenwood, and D. Wesloh, “Ogc R⃝sensorml: Model and xml encoding standard,” Technical Stan-dard, vol. 2, 2014.

[18] B. Campbell, J. Rosenberg, H. Schulzrinne, C. Huitema,and D. Gurle, “Session initiation protocol (sip) extension forinstant messaging”, rfc 3428,” 2002.

[19] (06/2017) Influxdb. [Online]. Available: https://home-assistant.io/components/influxdb/

[20] (06/2017) Grafana labs: The analytics platform for all yourmetrics. [Online]. Available: https://grafana.com/grafana

[21] Linphone open source voip project. [Online]. Available:http://www.linphone.org/

[22] (2017) Emergency context resolution with internettechnologies (ecrit). [Online]. Available: http://ecrit.sourceforge.net/lost server.html

[23] Q. T. Huynh, U. D. Nguyen, S. V. Tran, A. Nabili, and B. Q.Tran, “Fall detection system using combination accelerometerand gyroscope,” in Proc. of the Second Int. l Conf. onAdvances in Electronic Devices and Circuits (EDC 2013),2013.

[24] A. Van de Vel, K. Cuppens, B. Bonroy, M. Milosevic,K. Jansen, S. Van Huffel, B. Vanrumste, L. Lagae, andB. Ceulemans, “Non-eeg seizure-detection systems and po-tential sudep prevention: state of the art,” Seizure, vol. 22,no. 5, pp. 345–355, 2013.

[25] H. D. Masethe and M. A. Masethe, “Prediction of heartdisease using classification algorithms,” in Proceedings of theworld Congress on Engineering and computer Science, vol. 2,2014, p. 2224.